分布式水文模型区域分解并行计算方法及其应用

Domain Decomposition Parallel Computing Method of Distributed Hydrological Model and Its Application

针对区域分解并行计算方法中存在任务分配负载不均衡、通信时间长的问题,采用OpenMP编程模型,以WEP-L分布式水文模型汇流过程为例,设计了区域分解并行方法,首先通过子流域拓扑关系表确定子流域属于干流或支流,然后将每一条支流划分为一个水文区域,并保证水文区域中子流域按上游到下游的顺序排序,同时,为了实现分配的负载均衡,设计了基于贪心算法的优化调度,最后将支流任务分配给不同的线程计算,干流子流域采用串行计算。对黄河流域在不同线程条件下并行效果的测试结果表明,基于贪心调度的区域分解并行计算方法能有效提高汇流过程运算效率。

Aiming at the problems of unbalanced task distribution load and long communication time in the domain decomposition parallel computing method,taking the flow routing process of the WEP-L distributed hydrological model as an example,the parallel method of domain decomposition is designed using the OpenMP programming model.Firstly,whether the sub-watershed is a main stream or tributary is determined through the topological relation table of the subwatershed.Then each tributary is divided into a hydrological area,and ensure that the sub-watersheds in the hydrological area are sorted from upstream to downstream.In order to realize the load balance of distribution,the optimal scheduling is designed based on a greedy algorithm.Finally,the tributary tasks are assigned to different threads for calculation,and the serial computation is applied to the main stream sub-watershed.The parallel results under different thread conditions in the Yellow River basin are tested,and the results show that the parallel computing method of domain decomposition based on greedy scheduling can effectively improve the efficiency of flow routing process.